Object Detection and Segmentation Method for Unmanned Vehicles in Bird's-Eye View Based on Peripheral Camera

Abstract

Target detection based on peripheral vision cameras is a current research hotspot in the field of autonomous driving perception. This paper proposes a target detection and segmentation algorithm based on the surrounding view camera in the bird's-eye view. The algorithm combines the Cross-view Transformer with the road segmentation function and the target detection algorithm and uses a network to simultaneously realize the detection and segmentation of two tasks. Aiming at the problem of inaccurate angle regression when detecting objects under the bird's-eye view, this paper proposes a coding form based on direction vectors and a loss function based on cosine similarity. The test results show that the algorithm proposed in this paper has a good detection effect on the premise that it has little impact on the segmentation accuracy of the original network. The functions of road segmentation and object detection under the same backbone network are realized.